Arao AI
Back to search results

Qwen3.5-9B · Q2_K

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.40L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-9B (Q2_K) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

277.25 tok/s

Decode

Decode speed

1,119.33 tok/s

Prefill

Prefill speed

0.73 s

TTFT

Time to first token

12.5 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · Q2_K (current)Decode 277.25 · Prefill 1119.33 ·
  2. Ollama · Q5_K_M Decode 166 · Prefill 599.13 ·

Core figures

Decode (avg)
277.25 tok/s
Prefill (avg)
1,119.33 tok/s
TTFT (avg)
0.73 s
VRAM (avg)
12.5 GB
MTP acceptance rate
—
TTFB
2.22 GB
Power draw
424.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-9B
Quantization
Q2_K
Framework
llama.cpp
Version
v0.40
Context length
2048 tokens
Batch size
512
GPU layers
28
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
NVIDIA RTX 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
12.5 GB
OS
Ubuntu 24.04
Driver
565.57
CUDA
12.8
Power draw
424.7 W

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.40 · Ubuntu 24.04 · CUDA 12.8 · 2048 ctx

    277.25 tok/s

    Decode

    1,119.33 tok/s

    Prefill

    0.73 s

    TTFT

    12.5 GB

    VRAM

    —

    MTP

    424.7 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=13 running model_id=49 at Q2_K